hedging or market timing? selecting the interest rate exposure...

32
THE JOURNAL OF FINANCE VOL. LX, NO. 2 APRIL 2005 Hedging or Market Timing? Selecting the Interest Rate Exposure of Corporate Debt MICHAEL FAULKENDER ABSTRACT This paper examines whether firms are hedging or timing the market when selecting the interest rate exposure of their new debt issuances. I use a more accurate measure of the interest rate exposure chosen by firms by combining the initial exposure of newly issued debt securities with their use of interest rate swaps. The results indicate that the final interest rate exposure is largely driven by the slope of the yield curve at the time the debt is issued. These results suggest that interest rate risk management practices are primarily driven by speculation or myopia, not hedging considerations. WHEN FIRMS SELECT THE INTEREST RATE exposure of their liabilities and use deriva- tives to alter that exposure, are they hedging or timing the market? The em- pirical literature has attempted to estimate the sources of value creation stem- ming from hedging by examining the cross-sectional variation in the use of derivatives by firms (see Nance, Smith, and Smithson (1993), Mian (1996), and Graham and Rogers (2002), for instance). Implicit in most of these examinations is the assumption that firms use derivatives solely for the purpose of hedging. However, before we can estimate the value created by hedging through the se- lection of firms’ interest rate exposures, we must document that the choices of those exposures are for the purpose of hedging, and not attempts to reduce their cost of capital. This paper addresses that issue. Focusing on the selection of the interest rate exposure of newly issued corporate debt, and the use of in- terest rate swaps to modify that exposure, I examine whether firms really are hedging, or are instead timing the market in an attempt to reduce their cost of capital. If firms are hedging, then the choice of the interest rate exposure of the firm’s liabilities should be driven by the sensitivity of a firm’s cash f low to movements Michael Faulkender is from Olin School of Business, Washington University in St. Louis. This paper was previously entitled “Fixed versus Floating: Corporate Debt and Interest Rate Risk Management.” I would like to thank Greg Brown, Kent Daniel, Michael Fishman, Thomas Hazlett, Ravi Jagannathan, Elizabeth Keating, Todd Milbourn, Todd Pulvino, Rick Green, Stephen Sapp, an anonymous referee, and seminar participants at Northwestern University, Rice University, the University of Oklahoma, the University of Miami, the University of Michigan, the University of Minnesota, the University of Virginia, Washington University in St. Louis, and the Western Fi- nance Association Annual Conference for their valuable feedback. I also thank Marc Katz at Bank of America Securities for his helpful discussions, John Cochrane for generously providing data, and Eric Hovey for providing research assistance. I am especially indebted to Mitchell Petersen for his ongoing guidance. All errors are mine. 931

Upload: others

Post on 12-Jun-2020

3 views

Category:

Documents


0 download

TRANSCRIPT

THE JOURNAL OF FINANCE • VOL. LX, NO. 2 • APRIL 2005

Hedging or Market Timing? Selectingthe Interest Rate Exposure of Corporate Debt

MICHAEL FAULKENDER∗

ABSTRACT

This paper examines whether firms are hedging or timing the market when selectingthe interest rate exposure of their new debt issuances. I use a more accurate measureof the interest rate exposure chosen by firms by combining the initial exposure of newlyissued debt securities with their use of interest rate swaps. The results indicate thatthe final interest rate exposure is largely driven by the slope of the yield curve atthe time the debt is issued. These results suggest that interest rate risk managementpractices are primarily driven by speculation or myopia, not hedging considerations.

WHEN FIRMS SELECT THE INTEREST RATE exposure of their liabilities and use deriva-tives to alter that exposure, are they hedging or timing the market? The em-pirical literature has attempted to estimate the sources of value creation stem-ming from hedging by examining the cross-sectional variation in the use ofderivatives by firms (see Nance, Smith, and Smithson (1993), Mian (1996), andGraham and Rogers (2002), for instance). Implicit in most of these examinationsis the assumption that firms use derivatives solely for the purpose of hedging.However, before we can estimate the value created by hedging through the se-lection of firms’ interest rate exposures, we must document that the choicesof those exposures are for the purpose of hedging, and not attempts to reducetheir cost of capital. This paper addresses that issue. Focusing on the selectionof the interest rate exposure of newly issued corporate debt, and the use of in-terest rate swaps to modify that exposure, I examine whether firms really arehedging, or are instead timing the market in an attempt to reduce their cost ofcapital.

If firms are hedging, then the choice of the interest rate exposure of the firm’sliabilities should be driven by the sensitivity of a firm’s cash flow to movements

∗Michael Faulkender is from Olin School of Business, Washington University in St. Louis. Thispaper was previously entitled “Fixed versus Floating: Corporate Debt and Interest Rate RiskManagement.” I would like to thank Greg Brown, Kent Daniel, Michael Fishman, Thomas Hazlett,Ravi Jagannathan, Elizabeth Keating, Todd Milbourn, Todd Pulvino, Rick Green, Stephen Sapp,an anonymous referee, and seminar participants at Northwestern University, Rice University, theUniversity of Oklahoma, the University of Miami, the University of Michigan, the University ofMinnesota, the University of Virginia, Washington University in St. Louis, and the Western Fi-nance Association Annual Conference for their valuable feedback. I also thank Marc Katz at Bankof America Securities for his helpful discussions, John Cochrane for generously providing data, andEric Hovey for providing research assistance. I am especially indebted to Mitchell Petersen for hisongoing guidance. All errors are mine.

931

932 The Journal of Finance

in interest rates. By matching the interest rate exposure of the liabilities to thatof their assets, firms can reduce the variability of their cash flows. As a result,firms may lower their expected costs of financial distress (Smith and Stulz(1985)), as well as minimize how often they have to raise expensive externalcapital (Froot, Scharfstein, and Stein (1993)). Alternatively, if firms believe theycan time the market, thereby reducing their cost of capital, then the interestrate exposure selection should be driven by movements in interest rates. Firmsmay believe, as suggested in the Harvard Business School case study “Liabilitymanagement at General Motors” (Tufano (1995)), that they can reduce theirinterest costs by “actively managing” their interest rate exposure as interestrates change. When the yield curve is steep, firms that select a floating interestrate exposure will have significantly lower interest costs, at least in the shortterm, than firms with a fixed exposure. This paper seeks to determine whichof these objectives is the primary determinant of the ultimate interest rateexposure choice of firm liabilities.

A second critique of the existing empirical literature relates to the implicitassumption that firms that do not use derivatives are not hedging. A firm thatissues fixed rate debt has the same interest rate exposure, and therefore re-ceives the same theoretical benefits of smooth cash flows, as one that issuesfloating-rate debt and swaps it to a fixed rate. Yet, when the empirical risk man-agement literature has examined interest rate hedging, it has often equatedfirms that borrow floating and swap to a fixed interest rate exposure with be-ing hedgers, while the fixed rate debt users that do not swap were considerednonhedgers (Mian (1996), Nance et al. (1993)). As the above example illustrates,firms may be managing their risks, especially interest rate risk, by means otherthan derivatives usage. By issuing a debt contract that is correctly aligned withtheir desired interest rate exposure, there is no need to enter into an accompa-nying derivative contract. Thus, an empirical examination of whether firms arehedging, and what the benefits are of smooth cash flows, should examine thefinal interest rate exposure of the debt, not the intermediate step of how manyderivatives the firm uses. After all, the theoretical literature does not explorethe question of how firms achieve a particular exposure; it focuses on what fi-nal exposure a firm should have (see, for example, Smith and Stulz (1985) andFroot et al. (1993)).

Following this argument, the variable of interest in this paper is the finalinterest rate exposure of newly issued debt instruments. Modifications to theinterest rate exposure of debt securities via interest rate swaps are often madeat the time the debt is issued. This suggests that the final interest rate exposureof the new debt issue best reflects the implementation of the firm’s interestrate risk management program. To construct this measure, I collect data onboth bond issuances and bank loan originations, noting the initial interest rateexposure of the debt. These data are combined with hand-collected informationfrom firm annual reports on the interest rate swaps the firm entered as partof its financing. This methodology produces the final risk exposure, which iswhat the theoretical literature describes, and it makes possible an analysis ofhow firms choose to arrive at this exposure. Using this measure, I examine

Hedging or Market Timing? 933

whether the primary objective of the interest rate practices of firms is to hedgetheir cash flow or to reduce their cost of capital. An additional benefit of thismethodology, as opposed to the more traditional examination of derivativesusage, is that by looking at incremental issuances over a 6-year period, I amable to analyze time-series variation at monthly intervals, and not just examinecross-sectional variation.

The empirical results presented in this paper suggest that firms are mostlyinfluenced by market timing considerations. Firm-specific proxies for cash flowinterest rate sensitivity perform poorly in predicting the final exposure of thedebt. Contrary to the results that would be suggestive of hedging, firms withcash flows that are negatively correlated with interest rates are just as likelyto issue floating-rate debt as those with cash flows that are positively corre-lated with interest rates. These results are robust to the measure of interestrate sensitivity that is used. Instead, the strongest determinant in explainingfinal interest rate exposure is the yield spread, the difference in yields betweenlong-term and short-term bonds. As the yield curve steepens, firms are morelikely to take on floating-rate debt. Specifically, a one standard deviation in-crease in the spread between the yields of 10-year and 1-year Treasury bonds(44.5 basis points) increases the likelihood of observing a floating final interestrate exposure from 30% to 41%.

This result could be interpreted in several ways. Firms may be respondingto changes in market prices, as reflected by the yield spread. Alternatively, theyield spread may be a proxy for expectations of a recession, and firms are actu-ally responding to macroeconomic conditions. In separating these two effects,I find evidence that as the expected economic situation worsens, at both themacro level and the industry level, firms are more likely to choose a fixedrate exposure for their newly issued debt. This finding is consistent with firmsmeasuring risk at the industry level rather than at the firm level, suggest-ing that firms’ interest rate management may still be consistent with riskreduction. However, the yield spread is still a significant determinant of thechosen interest rate exposure, suggesting that firms are trying to lower theirshort-term cost of capital, even in the face of potential economic downturns. Inaddition, I examine whether the result is consistent for firms of different sizeand debt source, and find that small, bank-dependent firms are just as respon-sive to movements in the yield spread as large firms with access to the bondmarket.

The sensitivity of the interest rate exposure of newly issued debt to the yieldspread is consistent with a couple of possible interpretations. Firms may bespeculating by incorporating their views of anticipated interest rate movementsinto their interest rate exposure decision, resulting in a significant sensitivityto the yield spread, if such views are on average correlated with the shape of theyield curve. Second, this finding is consistent with firms managing short-termearnings via their interest expense by modifying their liability interest rate ex-posure when the difference in the current interest payment between fixed andfloating interest rates is large. In speaking with treasury employees at numer-ous industrial companies, I heard evidence for both of these interpretations.

934 The Journal of Finance

The remainder of the paper is organized as follows. Section I reviews theliterature on hedging and market timing that relates to this topic. Section IIdiscusses the empirical strategy for examining these alternatives. Section IIIprovides a description of the data used in the empirical investigation as wellas summary statistics. The empirical tests on the final exposure of the debtissuances and numerous robustness checks for alternative explanations arepresented in Section IV. Section V concludes.

I. Literature Review

The theoretical hedging literature is replete with accounts of the poten-tial benefits that arise from hedging firm cash flow. First, financial distresscosts provide a reason to hedge by reducing the probability of entering dis-tress and therefore of enduring such costs (Smith and Stulz (1985)). Hedgingcan also increase debt capacity, allowing firms to capture a greater tax shieldbenefit while maintaining the same or even reducing expected costs of finan-cial distress (Leland (1998)). Costly external finance, like that explored inMyers and Majluf (1984), creates a preference for internal cash over externalborrowings. Thus, hedging creates value if firms forgo positive NPV projectsless often when their cash flows are more stable and require fewer outsidecapital infusions (Froot et al. (1993)). In the presence of a convex or grad-uated tax schedule, firms can reduce their expected tax payments by mak-ing their earnings less volatile (Smith and Stulz). A final hedging justifica-tion (Stulz (1984)) is that if managers bear idiosyncratic risk through theircompensation, hedging reduces the volatility of their compensation, which isbeneficial to shareholders if expected compensation to managers is a loweramount.

Many early empirical examinations of hedging examine the use of derivatives,essentially equating their use with the desire of the firm to hedge. Some exam-ine whether firms use derivatives (Mian (1996), Nance et al. (1993), and Geczy,Minton, and Schrand (1997)), while others examined the extent of their us-age (Berkman and Bradbury (1996), Gay and Nam (1999), Howton and Perfect(1999)). One critique of this line of research is the assumption that the firmsthat do not use derivatives are not hedging (see Petersen and Thiagarajan(2000) and Graham and Rogers (2002), for instance). Such firms may not facethe risk a derivative product remedies or they may hedge using methods otherthan derivatives.

Further research focuses on firms that face a specific commodity exposure,thereby limiting the analysis to firms in which a particular hedgable risk waspresent. Such papers examine gold mining firms (Tufano (1996), Petersen andThiagarajan (2000)), oil and gas producers (Haushalter (2000), Geczy, Minton,and Schrand (1999), and thrifts (Schrand and Unal (1998)). The same idea hasbeen applied to currency hedging by controlling for the level of currency risk,as measured by the amount of foreign assets, sales, or income (Allayannis andOfek (2001), Graham and Rogers (2002)). Allayannis and Weston, for instance,find that firms with foreign sales that use foreign currency derivatives have

Hedging or Market Timing? 935

higher values than firms with similar foreign currency exposures that do nothedge that risk with derivatives.

Applying this idea to interest rate risk may be more complex, since identifyinga firm’s interest rate exposure is not so straightforward. Graham and Rogers(2002) address this by estimating the interest rate exposure of the firm’s cashflow prior to the incorporation of interest expense and earnings from deriva-tive transactions. They classify firms as having a hedgable interest rate risk iftheir estimated exposure is significantly different from zero and is not offsetby floating-rate debt, or if the firm does not have a significant interest rateexposure but does have floating-rate debt. Assuming that derivatives are usedto hedge, they then analyze the determinants of the level of derivatives usageamong those firms having a hedgable interest rate or foreign currency expo-sure, attempting to determine whether taxes affect the extent of derivativesusage. In this paper, I test whether firms use interest rate derivatives to hedgetheir interest rate exposure or if they are used for nonhedging purposes. There-fore, rather than using a firm’s cash flow interest rate exposure as Graham andRogers do to determine the existence of a hedgable risk, I use it as an explana-tory variable, that if the objective of the firm is to hedge its cash flow, shouldpredict the likelihood of choosing fixed versus floating-rate debt.

However, hedging may not be the only factor in a firm’s decision regardingthe interest rate exposure of their debt. Firms may also be timing the marketby responding to changes in macroeconomic conditions, in an attempt to reducetheir cost of capital.1 Baker and Wurgler (2000) find evidence that firms aretiming the equity market, increasing the percentage of total issues that areraised in the form of equity prior to unusually low equity market returns. Othersfind market timing effects in debt markets as well. Barclay and Smith (1995)find that outstanding debt maturity is negatively related to the yield spread,while Guedes and Opler (1996) find that the maturity of new public debt issuesis negatively related to the yield spread. Baker, Greenwood, and Wurgler (2003)find that changes in aggregate debt maturity are negatively correlated withsubsequent excess bond returns, consistent with firms trying to reduce theircost of capital through their maturity choice. Allayannis, Brown, and Klapper(2003) find that firms are more likely to borrow in a foreign currency as thedifference between LIBOR and local interest rates increases, taking on currencyrisk in an attempt to reduce the firm’s cost of capital. In the case “LiabilityManagement at General Motors,” GM states that the goal of their interest ratemanagement program is: “To actively manage the Central Office liabilities totake advantage of the cyclical nature and volatility of domestic interest ratesand shifts in the shape of the yield curve to reduce GM’s overall cost of funds”(Tufano (1995, p. 4)). Thus, both the empirical and clinical literatures suggestthat market conditions may influence the timing and characteristics of new

1 Market timing can be thought of as responding to issue-specific mispricings due to informationasymmetries as in Myers and Majluf (1984). Alternatively, market timing can refer to responsesto changing macroeconomic conditions that are correlated with the cost of capital, such as riskpremia or measures of risk that have cyclical components. This paper uses market timing in thelatter context.

936 The Journal of Finance

debt issues. This paper examines whether the selected interest rate exposureis also influenced by market prices.

II. Empirical Strategy

Firms face interest rate risk from two sources: the interest rate sensitivityof their assets and the sensitivity of their debt. Therefore, if firms considervolatile cash flows to be sufficiently costly, their risk management goal is tomatch the final exposure of their debt instruments to the interest rate sensi-tivity of their cash flows. This suggests that if the objective of a firm’s inter-est rate management policy is to hedge, then firms that have cash flows thatare positively correlated with interest rates should prefer interest paymentson their liabilities that float, whereas firms with cash flows that are eitheruncorrelated or negatively correlated with interest rates would prefer fixedinterest payments. Alternatively, if firms do not perceive significant benefitsfrom hedging the interest rate risk that they face, then they may have an al-ternative objective in choosing the interest rate exposure of their newly issueddebt. Specifically, they may be timing the market in an attempt to reduce theircost of capital. The goal of this paper is to determine which behavior firms areengaging in.

To separate these effects, I examine the final interest rate exposure of in-cremental debt issuances, which is defined as the combination of the initialexposure of the debt and firm’s use of swaps to alter the debt’s exposure. Thismethod improves upon the existing literature by constructing a panel seriesthat allows for both cross-sectional and time-series analyses. By looking at in-cremental issuances, I can specifically determine which debt instrument thefirm is modifying with the swap. Also, since firms choose the exposure at thetime of the debt issuance, examining that choice most closely reflects the strat-egy of the firm. Alternative analyses, which I also investigate as robustnesschecks, are to examine the level of interest rate exposure of firms at yearlyintervals, since derivative usage is disclosed annually, and to examine averageinterest rate exposures of firms over time. The benefit of looking at the individ-ual debt issues is that I can observe the month that the interest rate exposurewas chosen and the structure of interest rates during that month. Since in-terest rates can be volatile, measuring rates at the monthly level rather thanat the annual level should greatly improve the statistical power of the tests.Using the final interest rate exposure of the new issues, numerous potentialmeasures of hedging and market pricing, and a number of control variables, Iestimate probit regressions to examine whether firms are primarily concernedwith matching the exposure of their assets to their liabilities or whether theyare timing the market.

The primary measure of the interest rate exposure of the firm’s cash flowprior to raising debt funds is similar to that used by others in the litera-ture (Graham and Rogers (2002), Visvanathan (1998), Fenn, Post, and Sharpe(1996)). The methodology is described in Section III. If firms are hedging, thenthe results should be that firms that are more positively exposed to interest

Hedging or Market Timing? 937

rates (meaning their cash flow rises when interest rates rise) should choose tohave their interest payments positively correlated with interest rates. There-fore, as firm interest rate exposure increases, its preference for floating-ratedebt should increase.2 In addition to the primary measure of interest rate expo-sure broadly used in the literature, I modify that measure to verify that resultsare robust to the specification of interest rate exposure.

The other set of variables of interest measure macroeconomic conditions atthe time when the debt funds were raised. If firms are trying to reduce theirshort-term cost of capital in response to changing market prices, then theirinterest rate exposure should vary with changes in the term structure. Theyield spread is measured as the difference in the 10-year and 1-year yield on theU.S. Treasury bonds. I also use the credit spread as another measure of marketpricing conditions, where the credit spread is defined as the difference betweenthe average Baa corporate bond and Aaa corporate bond, as rated by Moody’s.Additionally, if a firm’s likelihood of distress changes over time with changes inthe economy, then the decision regarding the interest rate exposure of the firm’sdebt may change with the state of the economy. Therefore, I use the value of theIndex of Leading Indicators from the Conference Board, Inc. and the Universityof Michigan’s Survey Research Center to measure expectations regarding futuremacroeconomic activity at the time of the debt issue. I also control for theconditions of the chemical industry at the time of the debt issuance. In theirindustry survey of the chemical industry, Standard & Poor’s cites the FederalReserve Board Industrial Production Index as being an important measure ofindustry strength (O’Reilly (2000)). Thus, I also use the value of this index tomeasure whether current industry strength impacts a firm’s willingness to takeon interest rate risk.

Control variables include firm leverage, potential costs of financial distress,size, and profitability. Firms with less debt should be less concerned about incur-ring financial distress and therefore may be less concerned about the volatilityof their interest rate payment (as similarly argued by Tufano (1996), Geczy et al.(1999), and Haushalter (2000), among others). I examine research and develop-ment expenditures, advertising expenditures, and capital expenditures (eachstandardized by the sales of the firm) as measures of potential distress costs(following others, such as Graham and Rogers (2002), Geczy et al. (1997), andAllayannis and Ofek (2001)). As these measures increase, firms may becomemore concerned about interest rate fluctuations that may force such investmentexpenditures to be cut in times of distress, and therefore impact the desired in-terest rate exposure. Larger firms may be better able to manage risk, so firmsize may affect the interest rate choice of firms due to a difference in the abilityof firms to achieve a particular exposure. I also use the profit margin to testwhether profitability affects a firm’s decision whether to endure interest raterisk.

2 Theoretically, firms whose cash flow sensitivity to interest rates is negative should actuallyprefer an inverse floating rate debt instrument, if their objective is to hedge. However, I found noevidence in the data of firms choosing such an exposure.

938 The Journal of Finance

After presenting the initial regressions testing whether firms are hedgingor timing the market, I use alternative specifications to separate out differentinterpretations of the finding. Following these are numerous robustness checksto determine if the effects found for the entire sample are symmetric or if thefactors that motivate a particular exposure differ, based upon the source of thefunds and the relative size of the debt issue. Finally, I decompose the finalexposure into the source of funds, the initial exposure, and the use of swapsto determine at which stage firms consider the final exposure of their debt. Asdocumented later, the empirical findings are robust to all of these alternativespecifications.

III. Data and Summary Statistics

In order to test these theoretical implications, I build a data set of debt fund-ings for firms in the chemical industry for the period from 1994 to 1999. Ichoose this period because it is the first year that the Statement of FinancialAccounting Standards (SFAS) 119 required firms to report detailed informationregarding the direction of their derivatives positions. The chemical industry waschosen because it is a manufacturing industry and is presumably affected bychanges in interest rates. Second, firms in this industry are consistently rais-ing funds for ongoing investment purposes, allowing for a large enough samplesize. In addition, there are a number of subindustries within the chemicalsclassification that provide a rich heterogeneity in the interest rate exposure ofthe firms and the availability of investment opportunities.

The source of the data for these debt fundings is SDC Platinum for the bondissues and DealScan for the bank loans. The fundings are consolidated intoone observation if there are multiple loans or multiple bond issues becauseof different maturities issued in the same month, unless they have differentinitial interest rate exposures. Credit lines are dropped because the flexibilitycharacteristic of such a product complicates the comparisons with term loansor bond issues. 10-Ks in the EDGAR (Electronic Data Gathering, Analysis, andRetrieval system at the Securities and Exchange Commission (SEC)) databaseare used to verify the raising of the debt, to identify the initial interest rateexposure (fixed or floating), and to determine whether or not the firm swappedthe debt issue.3 Examples of disclosures made with respect to swaps in the10-Ks are provided in the Appendix. If the firms are not in COMPUSTAT orin EDGAR for the year the debt funds were raised, then the observation isdropped. Quarterly cash flow amounts, as well as annual income statement

3 In a few instances, only notional values were given, so it was determined whether or not theissue was swapped by looking at the change in the notional swap amount in that fiscal year forthe exposure that would be attached to the initial loan. For instance, if the loan was floating andthe change in the notional amount of the firm’s pay fixed, receive floating swaps for that year wasequal to or greater than the amount of the loan, then the loan was considered to have been swappedto fixed. If the direction of the notional amounts were not stated for the firm, the observation wasdropped.

Hedging or Market Timing? 939

and balance sheet information are obtained from COMPUSTAT.4 Using theseselection criteria, a sample of 275 debt issuances from 133 firms, spanning this6-year period, remains. The interest rate and macroeconomic data come fromDRI and the Federal Reserve Board.

For each observation, the quarterly cash flows for a 5-year period prior tothe debt issue are used to determine the sensitivity of the firm to changes inLIBOR. Since commercial debt is primarily tied to LIBOR if it is floating, theeffective choice for firms in my sample is to have fixed rate debt or floating-rate debt tied to LIBOR. If those are the choices, then it is the comovementwith LIBOR that is relevant for determining the hedging benefits.

Ideally, I would like to have an interest rate sensitivity measure for the po-tential future cash flow distribution. However, this is not readily available.Instead, the observed quarterly cash flows for the 5 years prior to the borrow-ings are used to estimate interest rate betas. Cash flow analysis allows for ameasure of interest rate exposure prior to the incorporation of interest expenseor derivatives usage that would alter the underlying exposure that the firm istrying to match. Specifically for the 5-year period prior to the raising of the debtfunds, using quarterly data, I run the following regression for each observation:

(Cash flowit/Book Assetsit) = α + βCF,i(LIBORt) + εit, (1)

where LIBORt is the average 6-month LIBOR interest rate during quarter t. Iuse βCF,i, which I refer to as the cash flow beta, as my measure of the firm’sinterest rate exposure at the time the debt funds were raised. In the regressionsusing βCF, I truncate the values at the 10th and 90th percentiles because a fewof the measured betas were outliers, due to the small number of observationsof quarterly cash flow prior to the issue.5

Table I provides summary information on the debt issuances in the sample,divided by the source of funds. Of the observations, 32% are funds raised frombanks. Of these, 90% were floating-rate loans, whereas only 7% of the nonbankdebt instruments were floating rate. There is a pronounced correlation betweenthe source of funds and the initial interest rate exposure. Banks predominatelylend to firms using floating rates while bond issuances are primarily fixed,resulting in a correlation coefficient of 82.4% between the source of funds andthe initial interest rate exposure.

The data also suggest that there are substantial differences in firm charac-teristics between the two sources of debt funds. Firms that raise public debtare more than 10 times larger than firms raising funds from banks on aver-age, when measured either by sales or by the market value of the firm. Thisdata (consistent with Cantillo and Wright (2000) and Faulkender and Petersen(2004)) support the theory that smaller, less transparent firms access banks for

4 Cash flow is defined as operating income before depreciation minus tax expense minus capitalexpenditures minus the change in current assets (reduced by cash holdings) plus the change incurrent liabilities (excluding long-term debt due within 1 year).

5 The coefficients are even smaller when the measure is not truncated and the statistical signif-icance of the coefficient is unchanged.

940 The Journal of Finance

Table ISummary Statistics by Debt Source

This table provides summary statistics on the interest rate exposure of the debt in the sample,separated by the source of the debt. It includes the initial exposure, the swap information, andthe final exposure statistics for the different sources. It also provides the means, and medians inparentheses, of the dollar value of sales (in millions), and the market value (in millions) of the firmcorresponding to the debt funds.

Bank Loans Bond Issues Total Number

Number 89 186 275Initially floating 80 13 94

89.9% 7.0% 34.2%Swapped to floating 0 12 12

0.0% 6.5% 4.4%Swapped to fixed 17 1 18

19.1% 0.5% 6.5%Floating after swap inclusion 63 24 87

70.8% 12.9% 31.6%Sales 559 6,826 4,798

(96) (3,865) (1,266)Natural log of sales 4.66 7.78 6.77

(4.56) (8.26) (7.14)Market value of assets 1,638 23,433 16,379

(270) (6,449) (2,442)Natural log of market value 5.79 8.70 7.76

of assets (5.60) (8.77) (7.80)

funds, whereas larger, more transparent firms tend to raise funds in the bondmarket.

Among the bank loans, 19% were swapped, all from floating to fixed. Thusafter incorporating the swaps, 71% of debt issues raised from banks ended uphaving a floating interest rate exposure and the remaining 29% were fixed. Inexamining the funds raised in the form of bonds, 7% were swapped from fixedto floating, and there was one issue swapped from floating to fixed. Followingincorporation of the swaps, the final interest rate exposure was floating for13% of the bond issues and was fixed for the remaining 87%. Overall, afterthe incorporation of swaps, 32% of the debt issuances are floating, as opposedto 34% of them that were initially floating. Thus, the use of swaps does notsubstantially change the fraction of debt that is floating during the sampleperiod; rather, it rearranges the interest exposure among firms across time.

Summary statistics of the firm-specific measures of potential hedging ben-efits, the market pricing variables, and the control variables are presented inTable II. For each of these variables, the mean, median, and standard deviationare provided for the entire sample and segmented by their final interest rateexposure. The final column lists the difference in the means and medians of thetwo samples and denotes the statistical significance of those differences.

These univariate results forecast the findings in the multivariate regressions.The average debt issue that ends with a fixed rate corresponds to a firm witha more positive cash flow interest rate exposure than the exposure for those

Hedging or Market Timing? 941

Table IISummary Statistics by Final Exposure

This table provides summary statistics on the hedging and market timing variables for the entiresample, and then separately by the final exposure of the debt instrument. The final exposure isdefined as the exposure of the debt instrument obtained by combining the initial exposure withany interest rate swaps used to modify that exposure. The mean, median, and standard deviationare provided for each of the variables. The last column is the difference in the mean and median,respectively, for the two groups. Cash flow interest rate beta is the beta from regressing the cor-responding firm’s quarterly cash flow on LIBOR (as outlined in Section IV). Yield spread is thedifference between the 10-year Treasury bond and the 1-year Treasury bond in the month the debtfunds were raised. Index of leading indicators is the value of that index during the month the debtfunds were raised. Leverage is defined as long-term debt over market value of the correspondingfirm. Profit margin is defined as operating income before depreciation divided by sales for thecorresponding firm during the year of the debt issue.

Entire Ending EndingSample Floating Fixed Difference

Cash flow interest rate betaMean 0.150 0.093 0.177 −0.083Median 0.019 0.031 0.017 0.013Standard deviation 1.774 2.058 1.632

Yield spread (basis points)Mean 67.0 76.9 62.4 14.5∗∗Median 58.0 78.0 53.0 25.0∗∗∗Standard deviation 44.8 45.1 44.0

Index of leading indicatorsMean 103.0 103.2 102.9 0.3Median 103.4 102.6 103.9 −1.3∗∗Standard deviation 3.6 3.3 3.7

LeverageMean 17.0% 17.2% 16.9% 0.3%Median 13.0% 11.8% 14.2% −2.4%∗Standard deviation 14.6% 17.1% 13.3%

Advertising-to-sales ratioMean 1.9% 1.3% 2.2% −0.9%Median 0.0% 0.0% 0.0% 0.0%Standard deviation 5.9% 4.0% 6.6%

Natural log of salesMean 6.77 5.60 7.32 −1.72∗∗∗Median 7.14 5.17 7.41 −2.24∗∗∗Standard deviation 2.30 2.59 1.92

Profit marginMean 12.4% 0.5% 18.0% −17.5%∗∗∗Median 17.3% 14.3% 18.9% −4.5%∗∗∗Standard deviation 22.3% 30.2% 14.6%

Number of observations 275 87 188

∗, ∗∗, and ∗∗∗ correspond to the differences being significant at 10%, 5%, and 1% respectively.

issues that end up floating, although the difference is statistically insignificant.This finding is in contrast to what we would expect if firms were hedging.Consistent with market timing, however, debt issues that end up floating areissued when there is a larger difference in the yield spread, evaluated at boththe mean and median, than issues that end fixed. Specifically, the median fixed

942 The Journal of Finance

0%

10%

20%

30%

40%

50%

Jan

1994

Jan

1995

Jan

1996

Jan

1997

Jan

1998

Jan

1999

Jan

2000

Date

Per

cen

t D

ebt

Issu

es F

loat

ing

af

ter

Sw

ap In

corp

ora

tio

n

0.00

0.50

1.00

1.50

2.00

2.50

Yie

ld S

pre

ad

PercentFloating

YieldSpread

Figure 1. Percentage of debt with floating exposure after swap incorporationrelative to the yield spread over time. This figure displays the patterns of the choice of in-terest rate exposure of newly issued debt in the chemicals industry (SIC 2800s) and the movementin the yield spread over the sample period 1994 through 1999. The bars correspond to the percent-age of debt that has a floating interest rate exposure after incorporating swaps used to modifythe exposure of the newly issued debt, with the units listed on the left side. The line representsthe average yield spread for each month, defined as the yield on 10-year Treasury bonds minusthe yield on 1-year Treasury bonds, with the units listed on the right side.

rate debt issue is done when the difference between the 10- and 1-year debt is53 basis points, as opposed to 78 basis points for the median debt issuancethat floats, a difference of roughly 40%. This univariate result can also be seenin Figure 1, which graphs the movement of the yield spread over the sampleperiod and the annual percentage of debt issued in the industry that endswith a floating interest rate exposure. In 1994, the average yield spread was177 basis points and 40% of the issuances had a final interest rate exposurethat was floating, whereas in 1998, the average yield spread was 21 basis pointsand only 20.3% of the issuances had a floating rate exposure.

IV. Empirical Findings

This section examines the empirical findings that result from testing the pre-dictions presented earlier. The first series of regressions examines the deter-minants of the final interest rate exposure of the incremental debt instrument.Finding that firms are responding to changes in the term structure, I thendecompose the yield spread to test multiple interpretations of that finding. Fol-lowing that, multiple robustness checks are performed to further substantiatethe finding. Finally, I examine the determinants of the individual steps firmstake to reach the final exposure to determine why different firms arrive at aparticular exposure through different means.

Hedging or Market Timing? 943

A. Determinants of the Final Interest Rate Exposure

It is the final interest rate exposure on their incremental debt issue, notthe use of a particular instrument, that represents the interest rate risk man-agement decision made by firms. Therefore, in analyzing whether the goal ofcorporate interest rate risk management is to hedge or to time the market, I usethe final interest rate exposure of new debt issues as the explanatory variable.Table III presents the results of such analysis.

The first thing to notice is that, as shown in column 1, the interest ratesensitivity of the firm’s cash flow does not predict whether the firm chooses afinal fixed or floating exposure on its debt security. If firms are hedging, thesign of the coefficient for this variable should be positive, indicating that theyare matching the exposure of their assets to the exposure of their debt. Instead,the estimated coefficient is slightly negative and statistically insignificant. Thisresult suggests that firms with cash flows that are positively correlated withinterest rates are no more likely to have floating-rate debt than those thatare negatively correlated. This finding is inconsistent with firms hedging theinterest rate exposure of their cash flows.

I also examine alternative measures of the cash flow sensitivity to interestrates to determine whether firms really are not matching the firm-specific cashflow interest rate sensitivity of their assets to their liabilities, or whether themeasure is merely misspecified. The second column uses a cash flow sensi-tivity measure that takes the value of the estimated cash flow beta when itis statistically significant and is zero when the estimate is insignificant. Thethird column uses a discrete measure of the cash flow beta, taking the value 1when the estimated sensitivity is significantly positive, −1 when significantlynegative, and is 0 otherwise. However, as the results show, neither of the speci-fications yields a statistically significant coefficient. These results suggest thathedging the firm-specific interest rate risk of their cash flows may not be aprimary objective of most of the firms in the sample.6

Instead, a key determinant of the final interest rate exposure appears tobe the yield spread. Firms are much more likely to have a floating-rate ex-posure when there is a steep yield curve than when the yield curve is rela-tively flat. Statistically, the coefficient of the yield spread variable is significantat better than 1%. Also, its magnitude implies a large economic significance.Evaluated at the means, a one standard deviation increase in the differencebetween the 10-year and 1-year government bond yields of 44.5 basis points in-creases the likelihood of having a floating final interest rate exposure from 30%to 41%.

6 In unreported regressions, I used estimates of interest rate sensitivity using quarterly dum-mies to capture any cyclicality that may be present. Moreover, I ran regressions in which capitalexpenditures were not subtracted, since firms may be engaging in capital expenditures only whenthere are sufficient internal funds, as opposed to a strategy of investing when they have goodprojects. In neither of these specifications was the coefficient on that variable significant, nor didit significantly alter the other estimated coefficients.

944 The Journal of Finance

Table IIIDeterminants of Final Interest Rate Exposure

This table presents the results of probit regressions where the dependent variable takes the value 1if the final exposure of the debt funds is floating, and is 0 otherwise. The final exposure is defined asthe exposure of the debt instrument, obtained by combining the initial exposure with any interestrate swaps used to modify that exposure. Free cash flow beta is the estimated exposure to LIBORof the corresponding firm’s quarterly free cash flow for the 5 years (subject to data availability)prior to the debt issue. Free cash flow is defined as operating income before depreciation minustaxes minus capital expenditures and minus the change in working capital. Free cash flow beta(significant exposures) is the same as for the previous measure when the estimated coefficient wassignificantly different from 0, and is 0 otherwise. Discrete free cash flow beta takes the value 0if the estimated free cash flow beta was positively significant, −1 if negatively significant, andis 0 otherwise. Credit spread is the difference between the BAA and AAA Corporate Debt in themonth the debt funds were raised. Yield spread is the difference between the 10-year Treasurybond and the 1-year Treasury bond in the month the debt funds were raised. Percent currentlyfloating is the percentage of the firm’s total debt at the end of the fiscal year prior to the debt issuethat had a floating interest rate exposure after accounting for any swaps the firm had entered,assuming that short-term debt has a floating interest rate exposure. The standard errors are Whiteheteroskedastic consistent (White (1980)), adjusted for clustering by company.

(1) (2) (3) (4) (5) (6)

Interest rate exposureFree cash flow beta −0.022 −0.044

0.051 0.062Free cash flow beta 0.000 −0.001

(significant exposures) 0.003 0.003Discrete free cash flow beta 0.006 0.005

0.151 0.165Percent currently floating 0.283 0.278 0.280

0.407 0.404 0.409Market timing

Credit spread 0.526 0.531 0.531 −0.037 −0.041 −0.0260.643 0.643 0.642 0.702 0.697 0.691

Yield spread 0.676∗∗∗ 0.683∗∗∗ 0.684∗∗∗ 0.582∗∗ 0.592∗∗ 0.597∗∗0.238 0.238 0.234 0.252 0.251 0.247

Control variablesLeverage 0.326 0.314 0.315 0.324 0.298 0.294

0.582 0.585 0.583 0.673 0.673 0.671Advertising over sales −2.783∗ −2.787∗ −2.790∗ −1.789 −1.852 −1.848

1.541 1.537 1.526 1.538 1.535 1.510CapEx over sales −0.390 −0.396 −0.394 −0.380 −0.384 −0.389

0.337 0.343 0.341 0.366 0.375 0.370R&D over sales 1.892 1.874 1.869 1.242 1.123 1.121

2.278 2.280 2.276 2.511 2.503 2.481Ln(Sales) −0.138∗∗ −0.133∗∗ −0.133∗∗ −0.130∗ −0.122∗ −0.119∗

0.059 0.060 0.059 0.069 0.071 0.070Profit margin −1.302∗∗ −1.358∗∗ −1.357∗∗ −1.832∗∗ −1.912∗∗ −1.943∗∗

0.664 0.664 0.650 0.845 0.841 0.822Constant −0.310 −0.345 −0.346 −0.056 −0.020 −0.004

0.616 0.618 0.615 0.812 0.814 0.812Pseudo R2 0.166 0.165 0.165 0.165 0.163 0.163

∗, ∗∗, and ∗∗∗ correspond to the coefficients being significant at 10%, 5%, and 1%, respectively.

Hedging or Market Timing? 945

This finding is extremely robust to the definition of the yield spread as well. Ihave also measured the yield spread as the difference in the 30-year Treasuriesrelative to the 1-year, the 5-year relative to the 1-year, and the 10-year relativeto the 2-year (regressions not reported). In all cases, this term had the samestrong statistical and economic significance.

A number of the control variables are also significant predictors of the finalexposure. As firms spend more on advertising as a percentage of their sales, theyare less likely to float. Such a finding may suggest that firms view interest ratevariability as costly, regardless of whether their cash flows move with interestrates or in the opposite direction of interest rates, and that the perceived costof that interest rate variability depends upon the type of assets they have. Ifa firm’s assets are particularly vulnerable to high costs of financial distress,specifically those assets associated with having high-advertising expenditures,they take on less floating-rate debt.

Larger and more profitable firms are more likely to have a fixed interestrate exposure, while smaller, less profitable firms are more likely to float. Thesize finding is consistent with the clientele effect illustrated in the summarystatistics contained in Table I. Smaller firms borrow from banks at floatinginterest rates, whereas large firms borrow in the bond market at fixed interestrates. The fact that size is significant suggests that firms may not fully overcomethe clientele effect, even in the presence of a swap market. Additionally, theprofit margin is significantly negative in all of the specifications. Consideringthat the yield spread was positive over the entire sample period, meaning thatfixed rate debt probably had a higher interest rate than did floating-rate debt,one interpretation of this result is that less profitable firms preferred the lessexpensive floating-rate debt and were more willing to take on the interest rateexposure associated with the lower cost financing than were more profitablefirms.

A natural criticism of the methodology above is that looking only at the in-terest rate exposure of new issues ignores the interest rate exposure of existingissues. If firms are in fact hedging, then the choice of the exposure of the newdebt issue depends upon the existing mix of fixed and floating debt they alreadyhave outstanding, including any modifications made using swaps to the exist-ing debt. For instance, firms may be managing their aggregate liability interestrate exposure toward a long-term average, issuing floating debt when their per-centage of floating-rate debt is low and vice versa. If firms are following sucha strategy, then upon including the existing percentage of floating-rate debtin the econometric specification, one would expect a negative and statisticallysignificant coefficient corresponding to the current floating-rate percentage. Inaddition, if firms are hedging, then after controlling for the current percentageof debt that is floating, we should find that firms with a positive cash flowexposure to interest rates are more likely to issue floating-rate debt than arethose with a negative cash flow exposure. Therefore, I have collected from thefirm’s 10-Ks the amount of their long-term debt that is floating at the end of thefiscal year prior to the debt issue. To that amount, I add the firm’s short-termdebt, essentially assuming that the interest rate exposure of short-term debt is

946 The Journal of Finance

f loating.7 I then subtract the notional value of pay fixed, receive floating swaps,and add the notional value of pay floating, receive fixed swaps to get a measureof the amount of floating-rate debt the firm has in its capital structure, prior tothe new debt issue. I then divide that amount by the total debt of the firm, toget a percentage of the firm’s capital structure comprised of floating-rate debt,before the new issue.8 The results of these regressions are in columns 4–6 inTable III.

Consistent with the earlier results, the coefficient on the three measures ofinterest rate exposure are still statistically insignificant. Even after controllingfor the current percentage of floating debt in their capital structure, hedging theinterest rate exposure of their cash flow does not appear to be a significant objec-tive in determining the interest rate exposure of corporate debt. Interestingly,the coefficient corresponding to the current percentage of floating-rate debt ispositive, contrary to the hypothesis, and statistically insignificant. The yieldspread retains its statistical and economic significance found in the previousregressions. This finding is inconsistent with firms managing their aggregateliability interest rate exposure toward a long-term average. Instead, the pos-itive coefficient suggests that firms that have previously issued floating-ratedebt continue to prefer that exposure for their new issues.

A further check on whether hedging is a motivation in the interest rate riskmanagement practices of firms is to regress the percentage of floating-rate debton the estimated interest rate exposure of the firm’s cash flows. Firms may beboth hedging their cash flows and timing the market by operating within apercentage range of floating-rate debt, where this range is determined by theunderlying interest rate exposure, whereas movements within this range arebased upon market prices. This would suggest that the percentage of all debtthat is floating would be positively correlated with the estimated interest rateexposure, while individual issuances may not show that correlation. Havingcontrolled for the current percentage of floating-rate debt, the new issuanceshould still show a significant coefficient if firms are indeed hedging. How-ever, directly measuring the mix prior to the issuance of the new debt securityprovides additional robustness to the evidence found so far. Therefore, suchregressions are presented in the first three columns of Table IV, using a tobitspecification where the bounds are truncated at zero and one. All measures ofinterest rates are for the month in which the fiscal year ended prior to the debtissuance and all control variables are for the previous fiscal year.

7 Since short-term debt is maturing within 1 year, the amount of short-term debt the firm hasis subject to interest rate fluctuations over the next fiscal year, justifying it being treated as hav-ing a floating interest rate exposure. In unreported regressions, I used only long-term debt inconstructing the measure of existing interest rate exposure and found similar results.

8 For a few observations in the sample, the firm’s capital structure had no debt prior to this newdebt issue. In addition, for some of the observations, the 10-K for the year prior to the debt issuewas not available on EDGAR, so a measure of the existing percentage of floating-rate debt wasunavailable. For the regressions in which the percentage of existing debt is included as a controlvariable, these observations are not included in the estimation. Thus, there are 256 observationsfor regressions that include this measure.

Hedging or Market Timing? 947

Table IVDeterminants of the Existing Interest Rate Exposure

This table presents the results of tobit regressions where the dependent variable is the percentageof the total debt that has a floating interest rate exposure after accounting for any swaps the firmhad entered, under the assumption that all short-term debt is floating. The first three columnsexamine the floating percentage at the end of the fiscal year prior to a debt issue. The last threecolumns examine the average floating percentage of each firm at the end of fiscal years prior toa debt issuance and the independent variables are averages of the stated variables over the fiscalyears prior to a debt issue. Since each firm is only represented once, there are 108 observations inthe specifications presented in columns 4–6. The upper truncation is at 1 and the lower truncationis at 0. Free cash flow beta is the estimated exposure to LIBOR of the corresponding firm’s quarterlyfree cash flow for the 5 years (subject to data availability) prior to the debt issue. Free cash flowis defined as operating income before depreciation minus taxes minus capital expenditures andminus the change in working capital. Free cash flow beta (significant exposures) is the same asfor the previous measure when the estimated coefficient was significantly different from 0, andis 0 otherwise. Discrete free cash flow beta takes the value 1 if the estimated free cash flow betawas positively significant, −1 if negatively significant, and is 0 otherwise. Credit spread is thedifference between the BAA and AAA Corporate Debt in the month of the fiscal year end prior tothe firm raising new debt. Yield spread is the difference between the 10-year Treasury bond andthe 1-year Treasury bond in the month of the fiscal year end prior to the firm raising new debt.

(1) (2) (3) (4) (5) (6)

Interest rate exposureFree cash flow beta 0.019 0.023

0.012 0.019Free cash flow beta 0.001 0.003

(significant exposures) 0.001 0.002Discrete free cash flow beta −0.001 −0.011

0.031 0.055Market timing

Credit spread 0.232∗ 0.260∗ 0.242∗ 0.435 0.502∗ 0.508∗0.136 0.138 0.137 0.265 0.259 0.265

Yield spread 0.174∗∗∗ 0.174∗∗∗ 0.171∗∗∗ 0.202∗∗ 0.215∗∗∗ 0.208∗∗0.038 0.038 0.038 0.082 0.082 0.083

Control variablesLeverage −0.279∗∗ −0.298∗∗∗ −0.281∗∗∗ −0.171 −0.218 −0.199

0.108 0.110 0.109 0.189 0.187 0.188Advertising over sales −0.492∗ −0.483∗ −0.489∗ −0.562 −0.730 −0.508

0.258 0.259 0.259 0.471 0.494 0.471CapEx over sales −0.081 −0.087 −0.091 −0.073 −0.086 −0.078

0.074 0.074 0.074 0.079 0.079 0.080R&D over sales −0.251∗ −0.258∗ −0.230 −0.388∗∗ −0.398∗∗ −0.339∗

0.145 0.148 0.145 0.195 0.195 0.196Ln(Sales) −0.041∗∗∗ −0.044∗∗∗ −0.046∗∗∗ −0.038∗∗ −0.039∗∗ −0.443∗∗∗

0.010 0.010 0.010 0.017 0.017 0.017Profit margin −0.236 −0.216 −0.180 −0.432∗ −0.434∗ −0.311

0.160 0.160 0.157 0.242 0.237 0.234Constant 0.697∗∗∗ 0.701∗∗∗ 0.717∗∗∗ 0.550∗∗ 0.528∗∗ 0.529∗∗

0.125 0.126 0.126 0.221 0.220 0.223Pseudo R2 0.309 0.301 0.297 0.263 0.269 0.249

∗, ∗∗, and ∗∗∗ correspond to the coefficients being significant at 10%, 5%, and 1%, respectively.

948 The Journal of Finance

The results of these regressions also suggest that the overall f loating per-centage is unaffected by the estimated interest rate exposure of the cash flows.Using the same three measures of interest rate exposure as in the regressionsof the interest rate exposure of the new debt issue, the coefficients on thesevariables are all statistically insignificant. Also consistent with the previousresults, the coefficient on the yield spread is significant in all three specifica-tions. In these regressions, the yield spread is the measure of the yield on the10-year Treasury bond minus the yield on the 1-year Treasury bond at the endof the month of the fiscal year ending prior to the debt issue. Economically, thiscoefficient suggests that a one standard deviation increase in the yield spreadcorresponds to an increase of 10% in the percentage of the firm’s debt that isfloating.

Finally, I take the average of the percentage of debt outstanding with afloating-rate exposure for each of the fiscal years prior to a debt issue for eachfirm and regress that average exposure on the averages of the correspondingindependent variables. Therefore, each firm is represented only once in eachof the specifications, presented in columns 4–6 of Table IV. This methodologyreduces the impact of firms that access debt markets more often, should suchfirms be more likely to time the debt market and less likely to hedge. Addi-tionally, an average measure of the mix of floating-rate debt in a firm’s capitalstructure may best measure the target interest rate exposure of firms, if theyhave targets and those targets are driven by the estimates of their cash flowsinterest rate exposure, arrived at using a methodology similar to the one Ihave used to estimate cash flow exposure. However, using averages minimizesthe impact of variables with considerable time-series variation. Therefore, weshould expect to find that the impact of the yield spread in these specificationsis diminished.

The estimated coefficients corresponding to the cash flow interest rate expo-sure variables remain statistically insignificant in these specifications as well.Firms that have an estimated interest rate exposure that is positive on averageare not more likely to have a higher average percentage of floating-rate debtthan those with an estimated exposure that is negative on average, as wouldbe expected if firms are matching the interest rate exposures of the liabilitiesto their assets. The yield spread retains its strong statistical significance andthe credit spread is marginally significant in these specifications, despite theexpected reduction in the power of time-series variables.

Regardless of how the final interest rate exposure is measured, the resultssuggest that the primary objective in selecting the interest rate exposure forcorporate liabilities is to time the interest rate market, not to match the interestrate sensitivity of the liabilities to that of the firm’s cash flows.

B. Interpretation of the Yield Spread Result

The finding that the yield spread predicts the choice of final interest rate ex-posure is consistent with a number of explanations. Some firms may be system-atically timing the market because a particular interest rate exposure offers

Hedging or Market Timing? 949

them a relatively lower cost of capital at the time they are raising the debtfunds (consistent with Baker et al. (2003) as well as with the survey results ofBodnar, Hayt, and Marston (1998) and Graham and Harvey (2001)). Such a set-ting would imply that the risk-return tradeoff of bearing interest rate volatilitychanges with movements of the yield spread. This tradeoff may change for tworeasons: the compensation for bearing interest rate risk changes over time orthe cost of bearing interest rate volatility varies over time. Alternatively, therisk-return tradeoff may be constant, but firms’ time preferences differ fromthose of the marginal investor, so that managers place greater emphasis onshort-term cash flow than do the shareholders. This section investigates thesealternatives by decomposing the elements of the yield spread and determiningto which components firms are responding.

Firms would indeed be lowering their risk-adjusted cost of capital if theirability to endure risk was constant over time, but the premium for taking oninterest rate risk was to vary.9 Firms would be more likely to take on interestrate risk when the premium was high and more likely to lay off that risk whenthe return for bearing that risk was low. In a setting with a time-varying riskpremium, some firms would alter their choice between fixed and floating asthe premium, and therefore the risk-return tradeoff, changes. If the slope ofthe term structure proxies for a time-varying risk premium, this type of settingmay explain why a firm’s choice of interest rate exposure is determined by theyield spread.

A second interpretation is that the yield curve is a proxy for a time-varyingcost of interest rate risk. Specifically, Estrella and Mishkin (1996) find thatthe yield spread is correlated with the likelihood of an economic recession.Flat or inverted yield curves may suggest a higher likelihood of an impendingrecession, whereas a steep yield curve suggests that there is a low likelihood ofan impending recession. Since the likelihood of distress is probably higher inrecessionary times than in nonrecessionary times, the cost of bearing interestrate volatility may be greater when the yield curve is flat than when it issteep. If volatile interest payments are especially costly during recessions, firmswould be more likely to lock in their interest payments when there is a smalldifference between long- and short-term interest rates (when a recession is morelikely) than when that difference is large (when a recession is less likely). If thecompensation for bearing interest rate risk remains constant, but a firm’s costof bearing interest rate risk changes with movements in the term structure,then some firms may respond to this yield spread change in the risk-returntradeoff by altering their choice of exposure.

Finally, firms may simply be moving their interest costs across time, with-out reducing their overall risk-adjusted cost of capital because managers have

9 The term risk-adjusted is used because the literature (Fama and French (1993), for instance)finds that historically, long-term bonds earn higher returns than do short-term bonds, suggestingthat long-term bonds have a systematic risk component. Therefore, floating rates are expected tobe lower, but as a result, the firm bears interest rate risk and the difference is the compensationfor that risk. Therefore, the term risk-adjusted cost of capital is used to recognize this difference.

950 The Journal of Finance

a different time preference than the market does or because they are specu-lating on interest rate movements. Increases in the yield curve occur becauseexpected short-term rates in the future have risen relative to current short-terminterest rates. Such a movement in interest rates would increase the differencebetween the current interest payments under floating rates versus under fixedrates. The firm can reduce its short-term interest expense by entering into afloating-rate debt instrument rather than fixing the rate, even though over thelife of the debt instrument, the expected risk-adjusted cost of capital remainsthe same. However, this is optimal from the perspective of a manager who has ahigher rate of time preference than the market does because the relative benefitof moving forward cash flows increases as the yield curve steepens. Similarly, ifmanagers take rate views that are correlated with the shape of the yield curveand this rate view is incorporated into the selection of interest rate exposure,then we may see a correlation between the yield curve and the interest rateexposure choice of firms.

To decompose the yield spread into its components, and to capture the effectsof a time-varying risk premium separate from the portion simply capturing thedifference between current and expected future interest rates, I use the singlefactor proposed by Cochrane and Piazzesi (2002) that has been found to covarywith bond risk premia. If firms are responding to time-series variation in thecompensation paid to take on interest rate risk, then the coefficient on the riskpremium variable should be statistically significant.

To capture variations in the cost of bearing interest rate risk over time, I usetwo measures of macroeconomic conditions. The first is the value of the Indexof Leading Economic Indicators at the time of the debt issue. The coefficienton this variable is expected to be positive if firms are less averse to locking inthe interest rate on their debt issue when economic conditions are expected tobe strong. As an alternative measure, I use the Industrial Production Indexfor the chemical manufacturing sector for the month in which the debt wasissued, as reported by the Federal Reserve Board. According to Standard andPoor’s Industry Survey on the chemical industry, this index is considered a“key industry ratio” that provides “an indication of industry production trends”(O’Reilly (2000, p. 27)). The coefficient on this variable is also expected to bepositive if firms are better able to bear interest rate volatility when the industryis performing strongly.

Adding measures of time-varying risk premia and expected macroeconomicconditions allows the yield spread term to capture the residual effects of move-ments in the term structure. Specifically, the coefficient on this variable shouldbe measuring responses to changes in future expected short-term interest rates.If firms are merely moving their interest payments across time without chang-ing their risk-adjusted cost of capital, then the coefficient on the yield spreadshould still be significant, even after adding the other controls. Such a find-ing would be consistent with the interpretation that interest rate exposuredecisions are driven by either myopic behavior or speculation on the part ofmanagers.

Interestingly, firms are not responding to the time-varying risk premia. Asshown by the results in Table V, the coefficient on this variable is insignificant

Hedging or Market Timing? 951

Table VFurther Examination of Final Interest Rate Exposure

This table presents the results of probit regressions where the dependent variable takes the value1 if the final exposure of the debt funds is floating, and is 0 otherwise. The final exposure is definedas the exposure of the debt instrument obtained by combining the initial exposure with any interestrate swaps used to modify that exposure. Cash flow beta is the estimated exposure to LIBOR ofthe corresponding firm’s quarterly free cash flow for the 5 years (subject to data availability) priorto the debt issue. Credit spread is the difference between the BAA and AAA Corporate Debt in themonth the debt funds were raised. Yield spread is the difference between the 10-year Treasury bondand the 1-year Treasury bond in the month the debt funds were raised. The Return forecastingfactor is a proxy for the expected bond risk premium during the month of the debt issue (Cochraneand Piazzesi (2002)). Index of leading indicators and chemical production index are the valuesof those indices during the month the debt funds were raised. The standard errors are Whiteheteroskedastic consistent (White (1980)), adjusted for clustering by company.

(1) (2) (3) (4) (5)

Interest rate exposureCash flow beta −0.019 −0.023 −0.026 −0.021 −0.023

0.051 0.051 0.052 0.051 0.051Market timing

Credit spread 0.430 −0.359 0.142 −0.432 0.0550.678 0.787 0.666 0.837 0.719

Yield spread 0.771∗∗∗ 0.949∗∗∗ 1.153∗∗∗ 1.024∗∗∗ 1.232∗∗∗0.290 0.250 0.277 0.312 0.340

Return forecasting factor −0.067 −0.057 −0.0590.124 0.125 0.125

Index of leading indicators 0.072∗∗ 0.071∗∗0.035 0.035

Chemical production index 0.080∗∗∗ 0.080∗∗∗0.029 0.029

Control variablesLeverage 0.354 0.100 0.019 0.124 0.043

0.597 0.567 0.568 0.579 0.579Advertising over sales −2.758∗ −2.829∗ −2.875∗ −2.801∗ −2.843∗

1.542 1.543 1.559 1.537 1.552CapEx over sales −0.386 −0.326 −0.370 −0.323 −0.366

0.338 0.347 0.346 0.348 0.346R&D over sales 1.997 1.301 1.075 1.403 1.178

2.355 2.065 2.074 2.145 2.153Ln(Sales) −0.141∗∗ −0.146∗∗∗ −0.151∗∗∗ −0.148∗∗∗ −0.153∗∗∗

0.059 0.057 0.057 0.057 0.058Profit margin −1.288∗ −1.252∗ −1.254∗ −1.239∗∗ −1.240∗

0.674 0.665 0.676 0.677 0.687Constant −0.005 −7.241∗∗ −8.097∗∗∗ −6.878∗∗ −7.768∗∗∗

0.802 3.445 2.885 3.348 2.753Pseudo R2 0.167 0.178 0.189 0.179 0.190

∗, ∗∗, and ∗∗∗ correspond to the coefficients being significant at 10%, 5%, and 1%, respectively.

in all of the specifications. This lack of significance is not due to multicollinear-ity, even though the correlation between the risk premium factor and the yieldspread is 63.8% over the sample period, as the coefficient on the yield spread re-mains significant. This finding indicates that all of the explanatory power comesfrom changes in the yield spread that are independent of the price of risk, as

952 The Journal of Finance

measured by Cochrane and Piazzesi (2002). This implies that managers’ choiceof interest rate exposure is not driven by changes in the compensation paid tofirms that bear interest rate risk.

Conversely, firms are responding to changes in expected macroeconomic con-ditions, consistent with the cost of bearing interest rate risk changing over time.As shown by the results in Table V, the coefficients on both of the additionalvariables measuring expected future economic conditions are statistically sig-nificant ( p-value < 0.05) in the respective specifications and their signs areas predicted. These findings suggest that as the market’s expectation of a re-cession declines or as chemical sector production increases, debt instrumentsare relatively more likely to be floating than when the likelihood of a recessionis high or when sector production is low. This finding of significance for botheconomic indicator variables lends support to an interpretation that the costof bearing interest rate risk changes over time, altering a firm’s risk-returntradeoff and for some, their choice of interest rate exposure. It is evidence thatfirms are managing macroeconomic and industry risk, rather than firm-specificinterest rate risk.

However, in all of the specifications, the coefficient on the yield spread re-mains positive and statistically significant ( p-value < 0.01). A one standarddeviation decrease in the yield spread (44.5 basis points) decreases the likeli-hood of having floating-rate debt from 28% to 16%. In contrast, a one standarddeviation decrease in the Index of Leading Indicators decreases the likelihoodof having floating-rate debt from 28% to 20%. The larger economic magnitudeand greater level of statistical significance of the yield spread coefficient pro-vide strong support for the interpretation that firms are timing the market,attempting to reduce their short-term cost of capital.

C. Robustness Checks

Still another interpretation of this finding is that the yield spread is pickingup movements in the level of interest rates. Firms may actually be floating wheninterest rates are at high nominal levels and going fixed when interest ratesare relatively low. To determine whether firms are responding to the spread orto the level, I regress the final interest rate exposure of the new debt issues onboth the spread and measures of the nominal level of interest rates. The resultsare presented in Table VI.

It appears that firms are in fact responding to the spread, not to the level ofinterest rates. In the first column, I include the yield on 1-year Treasuries andthe difference in the yield between the 10-year Treasury and the 1-year Trea-sury. In the second column, the 10-year Treasury yield and the yield spread areused. In both specifications, the coefficient on the yield spread is significantlydifferent from zero while the coefficient on the level of interest rates, capturedby the Treasury yield, is not statistically different from zero. Firms appear tochoose the security that has relatively lower interest costs at the time of the is-sue, fixed when the difference is small and floating when the difference is large.They do not appear to be basing their choice on the nominal level of interestrates.

Hedging or Market Timing? 953

Table VILevel of Interest Rates versus Yield Spread

This table presents the results of probit regressions where the dependent variable takes the value1 if the final exposure of the debt funds is floating, and is 0 otherwise. The final exposure is definedas the exposure of the debt instrument obtained by combining the initial exposure with any interestrate swaps used to modify that exposure. Cash flow beta is the estimated exposure to LIBOR ofthe corresponding firm’s quarterly free cash flow for the 5 years (subject to data availability) priorto the debt issue. Credit spread is the difference between the BAA and AAA Corporate Debt in themonth the debt funds were raised. Yield spread is the difference between the 10-year Treasurybond and the 1-year Treasury bond in the month the debt funds were raised. The 1-year Treasuryyield and the 10-year Treasury yield are the yields for those bonds during the month the debt fundswere raised. Chemical production index is the value of that index during the month the debt fundswere raised. The standard errors are White heteroskedastic consistent (White (1980)), adjusted forclustering by company.

(1) (2)

Interest rate exposureCash flow beta −0.030 −0.030

0.052 0.052Market timing

Credit spread 0.922 0.9220.886 0.886

Yield spread 1.129∗∗∗ 0.829∗∗0.275 0.361

1-year Treasury yield 0.2990.200

10-year Treasury yield 0.2990.200

Chemical production index 0.083∗∗∗ 0.083∗∗∗0.029 0.029

Control variablesLeverage −0.082 −0.082

0.584 0.584Advertising over sales −2.709∗ −2.709∗

1.544 1.544CapEx over sales −0.393 −0.393

0.346 0.346R&D over sales 0.725 0.725

2.153 2.153Ln(Sales) −0.150∗∗∗ −0.150∗∗∗

0.058 0.058Profit margin −1.331∗ −1.331∗

0.686 0.686Constant −10.506∗∗∗ −10.506∗∗∗

3.228 3.228Pseudo R2 0.194 0.194

∗, ∗∗, and ∗∗∗ correspond to the coefficients being significant at 10%, 5%, and 1%, respectively.

Having determined that firms are primarily responding to pricing conditions,specifically the yield spread, I now examine whether that response is robustto the source of the debt funds. Is the responsiveness to pricing conditions thesame for firms that borrow from banks as for those that issued their debt inthe bond market? Firms that borrow from different sources may have different

954 The Journal of Finance

risk management strategies, or their ability to manage risk may vary withtheir debt source. Since the firms that go to banks are smaller and more risky(Faulkender and Petersen (2004)), they may be less able to endure interest ratevariability and therefore respond differently to movements in the yield spread.

I examine this by separating the yield spread variable into two variables.One takes the value of the yield spread if the debt source is a bank, and iszero otherwise, and the second takes the value of the yield spread if the sourceis the bond market, and is zero otherwise. If firms are responding differentlyto pricing conditions because risk management practices are indeed correlatedwith debt source, then the coefficients on these two terms should differ. I alsoinclude debt source in the regression to ensure that any differences in the coef-ficients are due to different reactions to the yield spread, not due to the sourceof funds. In the second specification, I do the same separation for the Indus-trial Production Index for the chemical industry to determine whether firmsare also responding differently to current industry conditions. The results arepresented in Table VII.

The results in the first column show that both types of firms—those thatborrow from banks and those that issue in the bond market—are more likely tohave a floating interest rate exposure as the yield spread increases. Althoughboth coefficients are significantly different from zero, they are not significantlydifferent from each other ( p-value = 0.68). Thus there does appear to be sym-metry in responses to changes in market pricing, regardless of whether thefirm borrowed from banks or from the bond market. This finding of symmetryis evidence that swap costs are not prohibitively large, since the smaller, bank-dependent firms that should be more affected by these costs are not any lessresponsive to yield spread changes.

The second column of Table VII includes the separated measures of the In-dex of Industrial Production for the chemical industry as well. Both coefficientson the Index variable are positive, statistically significant, and are not sig-nificantly different from each other ( p-value = 0.97). As industry conditionsimprove, firms are more likely to have a floating interest rate exposure regard-less of whether they borrowed from a bank or from the bond market, suggestingthat firms are less able to withstand interest rate fluctuations when economicconditions are expected to be poor. Even after including these terms, the coeffi-cients on the yield spread terms are still significantly different from zero andnot from each other ( p-value = 0.80). These results suggest that the primarypurpose of the interest rate risk management policies of firms, independent ofthe source of their debt funds, is to time the interest rate market and to manageindustry-wide risk, not to hedge firm-specific interest rate exposure.

Another possibility is that firms are only timing the market when choosingthe interest rate exposure of their smaller debt issuances.10 Since timing inter-est rates can increase the firm’s risk, they may only be responding to marketprices when the issue size is small relative to the amount of debt the firm has.If that is the case, then the yield spread should only be a significant predictor ofthe observed interest rate exposures for those debt issuances that are relatively

10 I thank an anonymous referee for pointing out this possible alternative explanation.

Hedging or Market Timing? 955

small. To test this, I divide the sample based on the size of the issue relative tothe firm’s long-term debt. If the issue size as a percentage of the firm’s long-termdebt is larger than the median, it is considered to be a large issue; otherwiseit is considered a small issue. Following the procedure implemented with theyield spread and the source of funds, the first interaction variable measures theyield spread for those issuances that are relatively large (and is zero otherwise),

Table VIIYield Spread Effect, Conditional on Source and Amount

This table presents the results of probit regressions where the dependent variable takes the value1 if the final exposure of the debt funds is floating, and is 0 otherwise. The final exposure is definedas the exposure of the debt instrument obtained by combining the initial exposure with any interestrate swaps used to modify that exposure. Cash flow beta is the estimated exposure to LIBOR ofthe corresponding firm’s quarterly free cash flow for the 5 years (subject to data availability) priorto the debt issue. Credit spread is the difference between the BAA and AAA Corporate Debt in themonth the debt funds were raised. Yield spread is the difference between the 10-year Treasurybond and the 1-year Treasury bond in the month the debt funds were raised. Yield spread ∗ banktakes the value of the yield spread in the month the debt funds were raised for firms that borrowedthe funds from a bank, and is 0 otherwise. Yield spread ∗ bond takes the value of the yield spreadin the month the debt funds were raised for firms that issued bonds, and is 0 otherwise. Chemicalproduction ∗ bank and Chemical production ∗ bond are the values of that index during the monththe debt funds were raised if the source was a bank or bond, respectively, and are 0 otherwise.Yield spread ∗ small takes the value of the yield spread in the month the debt funds were raisedfor firms whose debt issue as a fraction of total long-term debt was smaller than the median of thatfraction for the sample, and is 0 otherwise. Yield spread ∗ large takes the value of the yield spreadin the month the debt funds were raised for firms whose debt issue as a fraction of total long-termdebt was larger than the median of that fraction for the sample, and is 0 otherwise. Debt source isbank takes the value 1 if the source of the debt funds was a bank, and is 0 otherwise. Large debtissue takes the value 1 if the debt issue as a fraction of the firm’s total long-term debt was largerthan the median of that fraction for the sample, and is 0 otherwise. The standard errors are Whiteheteroskedastic consistent (White (1980)), adjusted for clustering by company.

(1) (2) (3)

Interest rate exposureCash flow beta −0.046 −0.055 −0.026

0.060 0.062 0.051Market timing

Credit spread 0.032 −0.470 0.5110.704 0.683 0.642

Yield spread ∗ bank 0.879∗ 1.436∗∗0.488 0.604

Yield spread ∗ bond 0.636∗ 1.258∗∗∗0.344 0.370

Yield spread ∗ small 0.904∗∗0.394

Yield spread ∗ large 0.473∗∗0.240

Chemical production ∗ bank 0.105∗∗0.046

Chemical production ∗ bond 0.102∗∗∗0.034

(continued)

956 The Journal of Finance

Table VII—Continued

(1) (2) (3)

Control variablesLeverage 0.167 −0.256 0.312

0.705 0.698 0.592Advertising over sales −4.096∗∗ −4.197∗∗ −2.753∗

1.833 1.919 1.523CapEx over sales −0.236 −0.208 −0.384

0.382 0.410 0.336R&D over sales 1.500 0.250 1.862

2.342 2.205 2.358Ln(Sales) 0.154∗∗ 0.142∗ −0.150∗∗

0.075 0.076 0.066Profit margin −1.760∗∗ −1.713∗∗ −1.255∗

0.726 0.776 0.660Debt source is bank 1.870∗∗∗ 1.760

0.507 5.884Large debt issue 0.240

0.387Constant −2.571∗∗∗ −12.604∗∗∗ −0.361

0.688 3.290 0.686Pseudo R2 0.355 0.384 0.170

∗, ∗∗, and ∗∗∗ correspond to significance at 10%, 5%, and 1%, respectively.

and the second does the same for the relatively smaller issuances. The resultsare presented in column 3 of Table VII.

The two yield spread variables both have positive and significant coefficients,suggesting that firms alter the interest rate exposure of their debt instrumentsas the yield spread changes, regardless of whether the new issue is a largeor small percentage of the debt in their capital structure. Also, the differencebetween the coefficients is positive, as would be expected if firms were morelikely to time the market on their smaller issuances than their larger ones,but they are not significantly different from each other ( p-value = 0.335). Thissuggests that firms do not view market timing to be an overly risky strategy,since we should otherwise see that they are unwilling to take on this risk withlarger issuances.

D. Decomposing the Final Exposure

Firms respond to changes in the yield spread, but at which point in arriving attheir ending exposure is the yield spread taken into account? This section looksat various intermediate steps taken in arriving at the final exposure to answerthat question. Table VIII presents the results of analyzing the determinants ofthe source of debt funds, the initial exposure of those funds, and the directionof a swap if one was used.

The first column provides the results of examining from where firms raisetheir debt funds. Recall from Table I that there is a high correlation between the

Hedging or Market Timing? 957

Table VIIIDecomposition of Final Exposure

The first column in this table presents the results of a probit regression where the dependentvariable takes the value 1 if the debt funds were raised from a bank, and is 0 otherwise. The secondcolumn presents the results of a probit regression where the dependent variable takes the value 1if the initial exposure of the debt funds is floating, and is 0 otherwise. The third column presentsthe results of an ordered probit regression where the dependent variable takes the value 1 if theexposure of the debt funds was swapped from fixed to floating, −1 if the debt funds were swappedfrom floating to fixed, and is 0 if the debt funds were not swapped. Cash flow beta is the estimatedexposure to LIBOR of the corresponding firm’s quarterly free cash flow for the 5 years (subject todata availability) prior to the debt issue. Credit spread is the difference between the BAA and AAACorporate Debt in the month the debt funds were raised. Yield spread is the difference betweenthe 10-year Treasury bond and the 1-year Treasury bond in the month the debt funds were raised.Chemical production index is the value of that index during the month the debt funds were raised.The standard errors are White heteroskedastic consistent (White (1980)), adjusted for clusteringby company.

(1) (2) (3)Source of Initial Swap

Dependent Variable Funds Exposure Direction

Interest rate exposureCash flow beta 0.020 0.015 −0.049

0.045 0.043 0.055Market timing variables

Credit spread 1.480∗∗ 0.561 −0.6100.677 0.598 0.531

Yield spread 0.085 0.446∗∗ 1.015∗∗∗0.234 0.228 0.357

Chemical production index −0.017 0.046 0.0490.029 0.029 0.030

Control variablesLeverage 0.306 0.543 −1.045

0.566 0.559 0.758Advertising over sales 0.223 0.388 −3.409∗∗∗

1.697 1.192 1.294CapEx over sales −0.440 −0.086 −0.252

0.299 0.284 0.354R&D over sales −0.080 1.432 −0.783

2.180 2.198 2.109Ln(Sales) −0.489∗∗∗ −0.220∗∗∗ 0.126∗∗

0.068 0.058 0.052Profit margin −0.214 −0.751 −0.762

0.752 0.639 0.583Constant 3.197 −4.280

2.984 3.043Cut point 1 3.815

3.033Cut point 2 7.456

3.058Pseudo R2 0.388 0.189 0.133

∗, ∗∗, and ∗∗∗ correspond to significance at 10%, 5%, and 1%, respectively.

958 The Journal of Finance

source of funds and the initial interest rate exposure (ρ = 0.82). The findingssuggest that even despite that correlation, firms do not choose their source offunds based on the likely interest rate exposure they will receive from thatsource. Notice that the yield spread is not a significant determinant of wherethe funds were raised. Instead, the determinants are the credit spread andthe size of the firm. The statistical significance of the coefficient on the creditspread suggests that as riskier, lower rated corporate debt becomes more ex-pensive relative to highly rated debt, firms are more likely to borrow frombanks.

The finding that firm size, as measured by the log of sales, is a significantpredictor is consistent with the theoretical literature’s argument that banksare better at piercing the information asymmetry between the firm and itslenders or perhaps they better monitor firms than the market does (Diamond(1991)). Larger firms are likely to be more informationally transparent and aretherefore more likely to borrow from the public market, whereas smaller, moreopaque firms should be borrowing from banks. These results are also consistentwith the empirical findings of Cantillo and Wright (2000) and Faulkender andPetersen (2004), who find that as firm size increases, firms are less likely toborrow from banks. Evaluated at the means, the coefficient on sales suggeststhat a one standard deviation decrease in sales results in the likelihood of us-ing bank debt increasing from 23% to 65%. Thus, there appear to be segmentedlending markets, based on size, that drive the source of funds, and the determi-nants of the final interest rate exposure have little power to explain the sourceof funds.

The second column of Table VIII looks at the initial interest rate exposure ofthe debt instruments in the sample. While highly correlated with the source offunds, as Table I illustrated, there are some firms that are able to borrow frombanks at fixed interest rates and there are some firms that do issue floating-rate bonds. The coefficient on the yield spread is significantly positive in thisspecification, indicating that firms change the default exposure of the sourceof funds due to movements in the yield spread. So even though there is a highcorrelation between the source of funds and the initial exposure, the smalldifference between these measures accounts for a large change in the coefficienton the yield spread. Notice that moving from the first to the second column, thecoefficient on the yield spread has increased by a factor of five.

The third column examines the direction of a swap, using an ordered probitwhere the dependent variable takes the value 1 if swapped from fixed to float-ing, −1 if swapped from floating to fixed, and is 0 otherwise. The coefficienton the yield spread is statistically significant in this specification as well, sug-gesting that interest rate swaps are being used in response to movements inthe yield spread, not in attempts to hedge cash flow exposure. The combinationof these results indicates that there are segmented lending markets where thesource of funds is determined by the size of the firm. When firms find the defaultinterest rate exposure of that source to be sufficiently costly, as a function ofthe steepness of the term structure, some firms are able to alter that exposureas part of the debt contract, while others use swaps in order to achieve theirdesired exposure.

Hedging or Market Timing? 959

V. Conclusion

Firms respond to market pricing conditions in an effort to reduce their short-term cost of capital. I find strong evidence that the shape of the yield curve isa key determinant of whether a firm’s newly issued debt instruments are fixedor floating. As the yield curve steepens, firms are more likely to have floating-rate debt, reducing their short-term interest payment relative to locking in ata higher fixed rate. As the yield curve flattens, firms are more likely to raisedebt funds that have a fixed final interest rate exposure. In addition, I findthat as the expectation of a recession increases, firms are more likely to lockin the interest rate of their newly issued debt. This finding is consistent witha higher cost of interest rate volatility during economic downturns relative toeconomic booms. However, even after controlling for expected macroeconomicactivity and industry conditions, the yield spread remains a strong determinantof the selected interest rate exposure, suggesting that managers may be myopicand/or they are speculating.

I do not find evidence that firms are hedging, as firm-specific measures ofinterest rate sensitivity are not predictive of firms’ choices of their interest rateexposure. The evidence is mixed with respect to potential financial distresscosts influencing the exposure firms choose. Firms with higher advertisingto sales ratios are marginally more likely to have fixed rate debt. Additionally,profitable firms are more likely to issue debt with a fixed interest rate exposure.

The source of funds does not affect the responsiveness of firms to market tim-ing variables. While firms that borrow from banks are smaller, and arguablyface higher costs in accessing more complex financial securities, they are justas likely to respond to movements in the yield spread and to changing macroe-conomic conditions as firms that raise their funds from the bond market. Con-trary to the existing literature, these results suggest that managing risk isnot prohibitively complex or expensive. The data further suggest that thereare segmented lending markets based on firm size, where the initial interestrate exposure of the debt is highly correlated with the source of funds. Despitethis correlation, firms do not change their source of funds as the yield spreadchanges. Instead, some firms are able to alter the initial interest rate exposurethat is the default for that source of funds, while others use swaps to alter thatexposure.

These results naturally lead to the question of why firms engage in this type ofbehavior. In speaking with a number of individuals who work in corporate trea-sury departments, I heard evidence for both short-term earnings managementand speculation. Modifying the interest rate exposure of liabilities when thereis a large difference between fixed and floating rates allows firms to reducetheir short-term interest expense, and thus report higher quarterly earnings.This is rational on the part of managers if they have a higher rate of timepreference because their incentive contracts place greater emphasis on currentrather than on long-term performance, or perhaps because they are approach-ing retirement age. Alternatively, some managers appear to be guided by theirviews regarding anticipated movements of interest rates in the future and thatsuch movements are correlated with the shape of the yield curve. While it is

960 The Journal of Finance

questionable whether such actions actually increase shareholder value, my dis-cussions with corporate treasury employees, in conjunction with my empiricalfindings, suggest that some firms believe that they do.

Appendix

The following are excerpts taken from 10-Ks to demonstrate how it was deter-mined that a debt instrument was swapped. Most explicitly disclose if the debtissue was swapped. However, in a few instances, like in the second example,it was inferred based on the changes in the notional amount of swaps and theamount of the debt issue.

Engelhard Corp., 10-K filed March 27, 1997In 1996, in connection with the $150 million 7% Notes due 2001 and the$100 million 7.375% Notes due 2006, the Company entered into ten for-ward starting swaps. These swap agreements were based on amounts andmaturities which coincided with the debt agreements. These agreementswere closed concurrent with the debt issuance. The resulting impact onthe weighted-average interest rate for 1996 was not material.

In the following example, the firm issued $142 million in fixed rate debt duringthe 1997 fiscal year. The change in notional swap amounts for the category“Fixed to Variable” is $217.5 million. Since this change exceeded the amount ofthe issued debt, it was considered part of the increase and therefore I consideredthe debt to have been swapped.

Air Products & Chemicals Inc., 10-K405 filed December 11, 1997The table below illustrates the contract or notional (face) amounts out-standing, maturity dates, weighted average receive and pay rates as of theend of the fiscal year, and the net unrealized gain (loss) of interest rateswap agreements by type at 30 September 1997 and 1996. The notionalamounts are used to calculate contractual payments to be exchanged andare not generally actually paid or received, except for the currency swapcomponent of the contracts. The net unrealized gain (loss) on these agree-ments, which equals their fair value, is based on the relevant yield curveat the end of the fiscal year.

Notional (Millions of dollars) Amount Maturities

30 September 1997Fixed to Variable $461.0 1998–2007Variable to Variable 60.0 2000–2001Interest Rate/Currency 354.1 1998–2005

$875.130 September 1996

Fixed to Variable $243.5 1997–2005Variable to Fixed 54.0 1997Variable to Variable 60.0 2000–2001Interest Rate/Currency 273.6 1998–2005

$631.1

Hedging or Market Timing? 961

REFERENCES

Air Products and Chemicals, Inc., 1997, SEC filing, Form 10-K, Annual report, EDGAR database.Allayannis, Yiorgos, Gregory Brown, and Leora Klapper, 2003, Capital structure and financial risk:

Evidence from foreign debt use in East Asia, Journal of Finance 58, 2667–2709.Allayannis, Yiorgos, and Eli Ofek, 2001, Exchange rate exposure, hedging, and the use of foreign

currency derivatives, Journal of International Money and Finance 20, 273–296.Allayannis, Yiorgos, and James Weston, 2001, The use of foreign currency derivatives and firm

market value, Review of Financial Studies 14, 243–276.Baker, Malcolm, Robin Greenwood, and Jeffrey Wurgler, 2003, The maturity of debt issues and

predictable variation in bond returns, Journal of Financial Economics 70, 261–291.Baker, Malcolm, and Jeffrey Wurgler, 2000, The equity share in new issues and aggregate stock

returns, Journal of Finance 55, 2219–2257.Barclay, Michael, and Clifford Smith, 1995, The maturity structure of corporate debt, Journal of

Finance 50, 609–631.Berkman, Henk, and Michael Bradbury, 1996, Empirical evidence on the corporate use of deriva-

tives, Financial Management 25, 5–13.Bodnar, Gordon, Hayt Gregory, and Richard Marston, 1998, Wharton 1998 survey of risk manage-

ment by US non-financial firms, Financial Management 27, 70–91.Cantillo, Miguel, and Julian Wright, 2000, How do firms choose their lenders? An empirical inves-

tigation, Review of Financial Studies 13, 155–189.Cochrane, John, and Monika Piazzesi, 2002, Bond risk premia, NBER Working paper no. 9178.Diamond, Douglas, 1991, Monitoring and reputation: The choice between bank loans and directly

placed debt, Journal of Political Economy 99, 689–721.Engelhard Corp., 1997, SEC filing, Form 10-K, Annual report, EDGAR database.Estrella, Arturo, and Frederic Mishkin, 1996, The yield curve as a predictor of U.S. recessions,

Current Issues in Economics and Finance, Federal Reserve Bank of New York, Volume 2,Number 7, 1–6.

Fama, Eugene, and Kenneth French, 1993, Common risk factors in the returns on bonds and stocks,Journal of Financial Economics 33, 3–56.

Faulkender, Michael, and Mitchell Petersen, 2004, Does the source of capital affect capital struc-ture? Working paper, Northwestern University and Washington University in St. Louis.

Fenn, George, Mitch Post, and Steven Sharpe, 1996, Debt maturity and the use of interest ratederivatives by nonfinancial firms, Working paper, Capital Markets Section, Board of Governorsof the Federal Reserve.

Froot, Kenneth, David Scharfstein, and Jeremy Stein, 1993, Risk management: Coordinating cor-porate investment and financing policies, Journal of Finance 48, 1629–1648.

Gay, Gerald, and Jouahn Nam, 1999, The underinvestment problem and corporate derivatives use,Financial Management 27, 53–69.

Geczy, Christopher, Bernadette Minton, and Catherine Schrand, 1997, Why firms use currencyderivatives, Journal of Finance 52, 1323–1354.

Geczy, Christopher, Bernadette Minton, and Catherine Schrand, 1999, Choices among alternativerisk management strategies: Evidence from the natural gas industry, Working paper, Univer-sity of Pennsylvania.

Graham, John, and Campbell Harvey, 2001, The theory and practice of corporate finance: Evidencefrom the field, Journal of Financial Economics 60, 187–243.

Graham, John, and Daniel Rogers, 2002, Do firms hedge in response to tax incentives? Journal ofFinance 57, 815–839.

Guedes, Jose, and Tim Opler, 1996, The determinants of the maturity of corporate debt issues,Journal of Finance 51, 1809–1833.

Haushalter, G. David, 2000, Financing policy, basis risk, and corporate hedging: Evidence from oiland gas producers, Journal of Finance 55, 107–152.

Howton, Shawn, and Steven Perfect, 1999, Currency and interest-rate derivatives use in US firms,Financial Management 27, 111–121.

Leland, Hayne, 1998, Agency costs, risk management, and capital structure, Journal of Finance53, 1213–1243.

962 The Journal of Finance

Mian, Shehzad, 1996, Evidence on corporate hedging policy, Journal of Financial and QuantitativeAnalysis 31, 419–439.

Myers, Stewart, and Nicholas Majluf, 1984, Corporate financing and investment decisions whenfirms have information that investors do not have, Journal of Financial Economics 13, 187–222.

Nance, Deana, Clifford Smith, and Charles Smithson, 1993, On the determinants of corporatehedging, Journal of Finance 48, 267–284.

O’Reilly, Richard, 2000, Chemicals: Basic, Standard & Poor’s Industry Surveys (Standard & Poor’s,New York).

Petersen, Mitchell, and S. Ramu Thiagarajan, 2000, Risk management and hedging: With andwithout derivatives, Financial Management 29, 5–30.

Schrand, Catherine, and Haluk Unal, 1998, Hedging and coordinated risk management: Evidencefrom thrift conversions, Journal of Finance 53, 979–1013.

Smith, Clifford and Rene Stulz, 1985, The determinants of firm’s hedging policies, Journal ofFinancial and Quantitative Analysis 20, 391–405.

Stulz, Rene, 1984, Optimal hedging policies, Journal of Financial and Quantitative Analysis 19,127–140.

Tufano, Peter, 1995, Liability management at General Motors, Harvard Business School, Case no.9-293-123.

Tufano, Peter, 1996, Who manages risk? An empirical examination of risk-management practicesin the gold mining industry, Journal of Finance 51, 1097–1137.

Visvanathan, Gnanakumar, 1998, Who uses interest rate swaps? A cross-sectional analysis, Jour-nal of Accounting, Auditing, and Finance 13, 173–200.

White, Halbert, 1980, A heteroskedasticity-consistent covariance matrix estimator and a directtest for heteroskedasticity, Econometrica 48, 817–838.